JPH0524521B2 - - Google Patents

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to an abnormality handling technology for complex equipment such as Plato, and particularly for extremely complex and large-scale equipment such as thermal power generation Plato and nuclear power generation Plato. The present invention relates to a method of recommending to an operator, using a computer, an operating method suitable for the equipment forming the system and for dealing with an abnormal state of the equipment.

[Conventional technology]

Conventionally, this type of operation (driving) guidance method was “On-line Power Plant Alarm and
Disturbance Analysis System”, ERRI Report
As described in NP-1379, April 1980,
Cause-Consequence Tree
A method using CCT (hereinafter abbreviated as CCT) has been proposed. CCT connects the cause and effect relationships of various events that occur in a plant in a tree-like manner based on preliminary analysis results, etc., and in the case of relatively small and simple plants, it can be used to create operation guides in case of abnormalities. It is very effective when used. However, the causal relationships described in CCT simply show the progression of events that were assumed in advance.
The amount of time it takes for an event to spread is also fixed. Therefore, in complex, large-scale plants, it is extremely difficult to use CCT to accurately judge the progress of actual events, which vary greatly depending on the operational status, type and severity of events that occur, etc. In addition, CCT not only lacks flexibility because it describes causal relationships that are clearly connected in a tree-like manner for each assumed cause, but also requires large-scale and complex CCT in the case of complex plants. Since the amount of data becomes enormous, it becomes extremely difficult to create and maintain it.

On the other hand, “Consultation” by E.Shortliffe
Systems for Physicians”, Proceeding of
Canadian Soc.for Computational Studies of
Intelligence, University of Victoria, 1980
describes an interactive medical consultation system that uses a small knowledge base. Since a knowledge base accommodates each partial causal relationship as an independent rule, it has the advantage of being easier to create and maintain than a CCT.

It is thought that the knowledge engineering method used in the medical consultation system described above can, in principle, be applied to operational guidance in the event of an abnormality in a plant. In other words, a knowledge base that contains rules for each part of the causal relationship between various events that occur in the plant is stored in a storage device, and when an abnormality occurs in the plant, information regarding the state of the plant is input through dialogue with the operator. It is at least theoretically possible to construct a system that searches and combines these rules based on the or equipment,
That might be enough.

[Problem to be solved by the invention]

However, when applying the above-mentioned interactive consultation method to the operation guide of complex equipment such as large-scale plants and systems, the interaction between the operator and the computer is complicated and time-consuming. , how to associate information obtained from a large number of various measuring instruments with knowledge or rules stored in a knowledge base, and how to relate various physical quantities that change complexly over time to causal knowledge. Many questions need to be resolved, such as how to deal with

Furthermore, this interactive consultation method is not suitable as it is for online connection with plants and the like. In this method, the system follows rules to find the premises (such as symptoms) necessary for each conclusion (such as a disease name) to hold, and then determines whether or not the premises hold (such as a specific symptom). whether there is)
The procedure is to ask questions of the interlocutor. That is, this is a so-called rule-driven process that starts from a rule and searches for data that satisfies the rule, and it is not efficient to apply this as is to online processing.

Further, conventional guidance systems do not have the ability to self-examine whether operations obtained through repeated reasoning are reliable, for example, whether safety is maintained.

Therefore, the purpose of the present invention is to apply knowledge engineering methods to quickly and accurately identify the causes of abnormal conditions in complex equipment such as plants and systems online, without the help of operators. Moreover, it is an object of the present invention to provide an operation guide method that can efficiently estimate and recommend the optimal operation to deal with the estimation.

[Means to solve the problem]

The method of the present invention directly receives the output of a detection device installed in a target device through an input device of a computer, generates element state data in a predetermined format suitable for inference processing using rules, and , searches for a rule corresponding to this element state data, further searches for rules that form a chain of causal relationships from this rule, and selects the cause of the abnormal state of the target device from among the causes at the end of the chain. is selected, and then an operation to deal with the selected cause is determined using countermeasure operation information indicating operations to deal with the causes in association with the causes. Each of these plurality of rules associates the required state of the target device with a cause and its effect as a pair. The element state data generated from the detection device output may be only data representing an abnormal state of the target device.

Searching for rules corresponding to element state data obtained from this input information, and further searching for rules forming a chain of causal relationships from this rule.
In contrast to the rule-driven processing of the conventional interactive consultation described above, this method is data-driven processing that starts from data and searches for rules that are satisfied by the data.

To search for a rule, first perform forward inference based on the element state data, add the obtained element state to the first element state data, and then add the element state thus added and the first element state data, respectively. It is possible to perform backward inference based on the above.

The rule search procedure is also a process of estimating the element state that should be reached later due to the cause estimated at a certain stage, or furthermore, estimating the element state that is estimated in this way from the initial state. The process may include a process of comparing the element states represented by the element state data and selecting or discarding the causes estimated so far based on whether there is a contradiction between them. When estimating the element state to be reached, the target element state can be determined by forward inference using the rules described above. Furthermore, these processes may be repeated recursively.

For multiple causes estimated in this way, for each of them, compare the element state that should be reached as estimated in the above manner with the element state represented by the first element state data. These causes can be selected based on the degree of agreement between them.

In the process of determining an operation to deal with the selected cause, the element state that should be reached after executing the operation found from the countermeasure operation information is predicted according to the estimation procedure described above. good. Furthermore, the resulting element state caused by this predicted element state may be predicted by forward inference using the plurality of rules. Also,
Such processing may be repeated recursively.

When multiple operations are obtained, the element states predicted as described above as a result of each operation are evaluated in light of predetermined operating principles, and the optimal operation is determined based on the results. can be determined.

[Effect]

According to the present invention, plant data detected by the detection device is inputted into a computer online,
Inference using rules is performed at high speed without requiring operator intervention, and the cause of an abnormal state of the target device can be quickly and accurately estimated. Moreover,
Data-driven inference processing allows online processing to be performed efficiently.

In particular, if only the element state data that indicates the abnormal state of the target device is generated during input processing, it is sufficient to include only the element states that have a causal relationship with the abnormal operating state of the target device in the rule. Processing is also performed based only on element states indicating abnormal states of the target device. Therefore, the knowledge base will be relatively small and its creation and
Maintenance is easy and the inference process takes a relatively short time, so the cause can be estimated at extremely high speed.

Additionally, if forward inference is performed prior to backward inference, in addition to abnormal conditions detected directly from the plant data, there will also be abnormal conditions detected from the plant data that should have occurred as a result of (or in conjunction with) such conditions. Since all the states that have not occurred are identified and backward reasoning is performed on them as well, the cause can be inferred over a wide range without omission.

Furthermore, by inserting a process of estimating the element state that should be reached after each estimated cause, it is possible to determine what state the target device should be in at present if the estimation of the cause is correct. If the cause is determined and the cause is selected based on the presence or absence of a contradiction between it and the element state obtained from the input data, the targets of subsequent processing will be narrowed down and efficiency will be increased. For each cause selected in this way, if you repeat the process of estimating the cause, estimating the element state that should be reached due to it, and selecting the elements, you can efficiently estimate the root cause. can do.

The last remaining possible causes are selected based on the degree of agreement between the element states that would be reached due to them and the element state data obtained from the input, e.g. The highest value is determined as the most probable cause.

Furthermore, in the operation decision process, by predicting the element state that should be reached after the operation is executed, it is possible to check whether the operation based on inference is really reliable.

〔Example〕

In the following, one embodiment of the present invention in a boiling water nuclear reactor plant will be described in detail.

First, an overview of the equipment configuration of a boiling water reactor plant to which an embodiment of the present invention is applied will be explained with reference to FIG.

Steam generated in the core 52 in the reactor pressure vessel 51 is sent to the turbine 56 through the main steam pipe 63, and then condensed in the condenser 57 to become water. This water is supplied as cooling water into the reactor pressure vessel 51 through the water supply pipe 64. As illustrated, the water supply pipe 64 sequentially connects the condensate pump 58, the demineralizer 59, the water pumps 60A, 60B, 61A, and 61B, and the feed water heater 62 from the upstream side. Water supply pump 60A and 60
B is a turbine-driven water supply pump, 61A
and 61B are motor-driven water pumps.
The motor-driven water pumps 61A and 61B are
Although they are temporarily driven when the reactor is started and stopped, only the turbine-driven water pumps 60A and 60B are driven during normal operation of the reactor. Cooling water supplied into the reactor pressure vessel 51 is sent to the reactor core 52 through a jet pump 53 by driving a recirculation pump 54 provided in a recirculation system piping 55.

The water level gauge 65 detects the water level (reactor water level) 67 within the reactor pressure vessel 51 . The flow meter 66 is
The flow rate flowing through each jet pump 53 is detected. By summing up the flow rates flowing through all jet pumps 53, the flow rate of cooling water flowing within the core 52 is obtained. Process quantities such as reactor water level67 and jet pump flow rate measured by numerous detectors are as follows:
It is sent to the electronic computer 68 and the process input/output device 6
8A to the central processing unit 68B.
The electronic computer 68 further includes memory (internal memory and external memory) 68C. Central processing unit 6
The results processed by 8B are displayed on a cathode ray tube (abbreviated as CRT) 71 provided in the control panel 70.

In this example, when an abnormality occurs in such a nuclear reactor plant, the cause of the abnormality is estimated using knowledge engineering techniques, and the operator proposes an operation method to deal with the abnormality. Present it to
This is an attempt to bring an abnormal situation at a nuclear reactor plant under control. For this purpose, the atomic computer 68
The memory 68C stores the cause/effect database 8,
Transition prediction database 11, operation database 1
2. Detailed database 9, case database 10
and a processing program 77.

Next, details of this embodiment will be explained with reference to FIG. 2 and the following figures. FIG. 2 shows the entire process carried out by the computer 68 in this embodiment in the form of a block diagram showing the functions working at each processing stage and the interconnections of the databases used therein.
Note that identical functionality or databases that work or are used in different processing stages are illustrated separately in each processing stage.

In this embodiment, the following five databases are used.

(1) The cause-and-effect database 8 is a database that stores data representing cause-and-effect relationships, consisting of combinations of causes and effects that directly result from those causes, regarding various events that may occur in a nuclear reactor plant. Such data corresponds to what is commonly called "rules" in the field of knowledge engineering.

FIG. 3 shows an example of the contents of this cause/effect database 8. Here, to simplify the drawing, both the cause part and the effect part are shown without distinguishing between attributes and their values, but in reality, as with normal rules, each part is divided into attributes and values. It is made up of the corresponding values.
For example, "void increase" indicates that the attribute "void amount" takes a value of "increase".

(2) The transition prediction database 11 is a database that stores data for predicting temporal changes in process quantities. This includes information on the operating status of each device, the status of each process quantity, etc., and the method for calculating the time required for the value to change to a certain value for the process quantity whose status is defined corresponding to the value. A method for calculating the value after a certain period of time is stored. FIG. 4 shows an example of the contents of the transition prediction database 11.

(3) The operation database 12 is a database containing operation plans under various conditions, and includes a condition part indicating a combination of conditions such as the operation status of each device and the status of each process amount, and a condition part that shows a combination of conditions such as the operation status of each device and the status of each process amount, It consists of an operation plan indicating possible operations. Figure 5 shows the operation database 1
An example of the contents of item 2 is shown below.

(4) The detailed database 9 is a database that stores detailed operating methods and operating conditions for each device. FIG. 6 shows an example of the contents of the detailed database 9.

(5) The case database 10 is a database containing preliminary analysis results and past operation records for various cases. FIG. 7 shows an example of the contents of the case database 10.

On the other hand, the processing program 77 has the following 11 types of functions as basic processing functions in this embodiment.

(1) Data conversion function 1 inputs plant data 19 obtained from the existing data collection system in the plant, transforms it into a form that can be handled within the operation guide device (A/D conversion, etc.), and configures a multiplex system. Regarding the values of each data obtained from multiple measuring instruments, majority vote etc. is used to narrow down the values to one according to decision logic, identify data that indicates an abnormality, and determine what to do for each data that indicates an abnormality. Identifier of whether it is data (e.g. data number) and value required for subsequent processing obtained by pre-processing (e.g. if water level, the amount of change from the previous measured value, the current water level and amount of change) symbols indicating each rating) are added to create element status data.

In this specification, data representing such individual partial phenomena is called element state data (abbreviated as element state), and a group of element states that collectively represent one state of the plant is called plant state data (plant state data). (abbreviated as state).

(2) The condition understanding function 2 is a function that inputs a certain plant condition and searches out all the results caused by this plant condition by forward reasoning based on the cause/effect database 8. That is, this function compares each element state in the input blunt state with the cause part of the cause/effect database 8, and if there is a match, the content of the result part corresponding to the cause part is added to the input state. A new element state is added to the previously obtained plant state, and the above process is repeated recursively to search until a possible final result is obtained, and the obtained final plant state is output.

The plant state 21 generated by the state grasping function 2 from the output of the data conversion function 1 is used as a reference plant state in several subsequent processes.

(3) The cause enumeration function 13 is a function that inputs a certain plant state and uses backward reasoning based on the cause/effect database 8 to find events that can be the direct cause of each element state within this plant state. . That is, this function compares each element state in the input plant state with the result part of the cause/effect database 8, and if there is a match, outputs the contents of the cause part corresponding to the result part.

(4) The prediction function 15 inputs a certain plant state and calculates the value of each element state within this plant state from the current state value (normal value when determining the cause, input value when determining the optimal operation). The time required for the change to reach a predetermined value is determined by executing a time calculation method stored in the transition prediction database 11. Next, the shortest time among the times obtained in this way is found, and the value that each element state of the input plant state can take after the elapse of this shortest time is calculated using the state calculation stored in the transition prediction database 11. Find it by executing the method. The various element states obtained in this way are added to the input plant state and output as the plant state at the next point in time.

(5) The consistency check function 16 compares several input plant states with the reference plant state 21 and checks if any of the former includes element states that are inconsistent with the reference plant state. For example, that plant state is removed and only the plant state that is consistent with the reference plant state is output.

(6) The determination function 18 is a function of selecting the plant state closest to the reference plant state 21 from among the several plant states output from the consistency check function 16.

(7) The operation enumeration function 14 inputs a certain plant state and searches for possible operation plans to deal with the plant state by comparing it with the condition section of the operation database 12. This is a function that adds and outputs the input plant status.

(8) The decision function 17 inputs several plant states including operation plans and element states expected as a result of the operation, selects the operation plan that produces the most desirable plant state as the operation result, and selects the operation plan that will produce the most desirable plant state as the operation result. This function outputs the plant status including the plan.

(9) The detailing function 5 inputs the plant state that is the output of the determination mechanism 17, searches for an element state in this plant state that means an operation of the equipment, and stores information about this operation in the detailed database 5. After confirming the operating conditions shown in , add detailed operating procedures to the input plant status. However, if the operating conditions are not satisfied, re-execution of the optimum operation determining function 4 is requested.

(10) The similar case search function 6 inputs the plant status output from the detailing function 5 and searches the case database 10 for causes and keywords (see Figure 7).
If there is a case where the cause matches the element state of the input plant state or the keyword matches with it at a certain rate or more, it is selected as a similar case and the identification information of the case (for example, case number) to the input plant status.

(11) The guidance creation function 7 is a function that inputs the plant status output from the similar case search function 6, formats it for display, and outputs it to a display device.

The cause determination function 3 and the optimal operation determination function 4 operate by combining some of the basic processing functions described above as follows.

The cause determination function 3 inputs the standard plant state 21 created by the state understanding function 2 from the output of the data conversion function 1, and checks the consistency with the cause enumeration function 13, the state understanding function 2, and the prediction function 15. The function 16 is activated repeatedly to find all causes that do not conflict with the reference plant state 21, and then the determination function 18 is activated to estimate the most probable cause of the plant abnormality. Then, the plant status 22 to which this cause is added is output.

The optimal operation determination function 4 inputs the plant state 22 which is the output of the cause determination function 3, and performs operations such as an operation enumeration function 14, a state understanding function 2, a prediction function 15,
The determining function 17 is activated repeatedly to determine the optimal operating method, and output the plant state 23 to which the operating method and the result of executing the operating method are added.

Each of the above processing functions is shown in more detail in FIGS. 8 to 20, and will be explained in detail in the description of a specific example below.

Next, this embodiment will be explained in more detail using a specific example.

For the sake of explanation, this is a phenomenon that is unlikely to occur in practice, but during the operation of a boiling water nuclear power plant, the shaft of the recirculation pump 54 that supplies cooling water to the reactor core becomes stuck, and if left untreated, it may occur. It is assumed that a situation occurs in which the reactor water level reaches L8, resulting in a scram, and when the reactor water level reaches L7 before the scram, an operator notices an abnormality and activates the operation guide device.

When the shaft of the recirculation pump becomes stuck, the flow rate of cooling water decreases and the amount of voids in the core increases. An increase in the amount of voids leads to a rise in the reactor water level. However, in reality, the phenomenon of shaft sticking and increase in void volume cannot be directly determined, and only the measured process variables, such as jet pump flow rate and reactor water level, are known.

When the operation guide device is activated by an operator's instruction, a plurality of plant data 19 measured at the nuclear power plant are input to the data conversion function 1 shown in FIG. 8. Data conversion function 1 is AD
Among the converted plant data, those that do not satisfy (exceed or fall short of) the set values are detected, subjected to predetermined preprocessing, and converted into plant status 20. In addition, in boiling water nuclear power plants, multiple measurement mechanisms are installed to measure important process quantities such as reactor water level as a countermeasure against failures, and their outputs are subjected to appropriate processing such as majority vote calculation. Then, the most probable measured value is input to the driving guide device.

FIG. 21 shows an example of the plant state 20 obtained in this way. In the figure, various data are expressed in human language to make it easier to understand, but in reality, they are written in computer processing codes (e.g. EBCDIC
Code) and format.

Plant status 2 which is the output of data conversion function 1
0 is input to the state grasping function 2 shown in FIG. The status grasp function 2 uses the input plant status 2.
If there is any information missing in 0, this missing information is supplemented by forward inference based on the cause/effect database 8 shown in FIG. That is, the state grasping function 2 searches for a cause part that matches each element state in the input plant state 20, and converts the contents of the result part corresponding to the matched cause part into the plant state 2.
Add to 0. This process is executed recursively, and when no more matching cause parts can be found, the plant state obtained up to that point (i.e., the element state found as described above is changed to the original plant state) 20) is output as the plant state 21.

However, in this case, there is no element state added by the above search, and the 22nd
As shown in the figure, a plant state 21 that is the same as the plant state 20 shown in FIG. 21 is output.
This is the plant state that will be the standard for some processes described later.

Next, the cause determination function 3 is activated. As shown in FIG. 10, the cause determination function 3 receives as input the plant status 21 output from the status grasping function 2, and performs a cause enumeration function 13, a status grasping function 2, a prediction function 15, and a consistency checking function 16. are activated repeatedly in this order to find all possible causes, and then, from among them, the determination function 18 determines:
Select the most probable cause 22.

First, the cause enumeration function 13 shown in FIG. 11 inputs the plant state 21, compares each element state therein with the result part of the cause/effect database 8 (Fig. 3), and selects the matched result part. Extract the corresponding cause part (that is, perform backward inference), and combine each obtained cause part into one plant state 21A.
Output as . In this case, for the sake of explanation, we assume that the identified causative factors are "increase in voids" and "increase in water supply flow rate."
As shown in the figure, "increase in voids" and "increase in water supply flow rate"
are each output as one plant state 21A. These plant states 21A are input to the state grasping function 2.

The state grasping function 2 (FIG. 9) uses recursive forward reasoning as described above to find all possible results from the input plant state. In this case, the plant status 21A shown in Figs. 23 and 24 is input, and the results caused by "increase in voids" and "increase in water supply flow rate" are determined by the cause and
The result database 8 is searched. In either case, "reactor water level rise" is obtained as a result, and this is added to each plant state 21B, that is, a plant state consisting of "void increase" and "reactor water level rise", "feed water flow rate increase" and " The plant state consisting of "reactor water level rise" (not shown) is predicted by prediction function 15.
is input.

The prediction function 15 predicts the transition of each plant state when the cause found as described above occurs, that is, when the number of voids increases and when the water supply flow rate increases. That is, prediction function 1
5 refers to the transition prediction database 11 (Fig. 4), and calculates the change time of the corresponding process quantity for each element state that can be predicted in each input plant state 21B, as shown in Fig. 12. Then, for each plant state, the minimum one of the obtained change times is determined.

In the current example, the change time for "reactor water level rise" for each of two cases, "void increase" and "feed water flow rate increase", is calculated using the time calculation method in the transition prediction database 11 shown in FIG. Calculate according to. However, when determining the cause, the "normal value" in the transition prediction database is used as the "current state value", and the value previously associated with the assumed cause is used as the "change value".
As a boundary value, from among the boundary values of "state values" recorded in the transition prediction database 11, if the change value is positive, various boundary values that are larger than the "current state value" (here, "normal value") are used. The minimum value among these may be used, and if it is negative, the maximum value among the various boundary values smaller than the "current state value" may be used.

Thus, in cases where "void increase" is the cause, the change in reactor water level is an average value, so
200mm/sec is used, the normal value L4 value 1000mm is used as the current state value, and since L5 and L6 are undefined as the state boundary value, the lower boundary of the recently defined upper state L7 Value 1300mm
is used. Therefore, the reactor water level change time is (1300-1000)/200=1.5, or 1.5 seconds.

Such calculations are performed for all element states for which predictive calculations are possible. In this case, since a calculation method directly corresponding to the element state "void increase" is not specified, the above calculation is not performed and only the reactor water level is calculated. Since there are no other element states input, the above 1.5 seconds is "void increase"
becomes the minimum change time corresponding to .

On the other hand, even in the case where the cause is "increase in water supply flow rate," the change in reactor water level is 200 mm/sec, the current state value (normal value) is 1000 mm, and the state boundary value is 1300 mm. The time is (1300-1000)/200=1.5, or 1.5 seconds, and since there are no other element states input, this 1.5 seconds is the "increase in water supply flow rate".
becomes the minimum change time corresponding to .

Next, the prediction function 15 calculates each element state after the elapse of the minimum change time determined as described above for each input plant state using the state calculation method in the transition prediction database 11,
The obtained element states are added to each input plant state to output a plant state 21C shown in FIGS. 25 and 26.

In this case, the reactor water level in the case of "void increase" is 1000 (normal value as current state value) + 200 (change value) x 1.5 (minimum change time) = 1300, that is, L7 is added as the reactor water level. (Figure 25). This calculation is also performed for all element states for which predictive calculation is possible. In this example, only the reactor water level is calculated, and a calculation method directly corresponding to the element state "void increase" is not specified, so the above calculation is not performed. Similarly, for the reactor water level in the case of "increase in water supply flow rate", 1000+200×1.5=1300, that is, L7, is added as the reactor water level (Figure 26).

Several plant states 21C obtained as described above are input to the consistency checking function 16. The consistency check function 16 checks the element states in each input plant state and the reference plant state (plant state 2) according to the procedure shown in FIG.
1) Check whether there are any inconsistencies among the element states in the process, and select only the plant states for which it is confirmed that there are no inconsistencies as the output 21D. A blunt state 21C that includes an element state inconsistent with an element state in the reference plant state 21 is discarded. For example, even though there is an "increase in voids" in the reference plant state 21, the input plant state 2
If 1C contains "void reduction", it will not be output. In this case, Figure 25 and 2
Assuming that there are no contradictions in the plant status shown in FIG. 6, the output will be 21D.

Next, the causes of each of the causes estimated as described above are further estimated. This is done by using the output 21D of the consistency checking function 16 as input and recursively calling the cause determination function 3. In other words, in this case, cause enumeration function 1
3 receives the plant status shown in Fig. 25 and Fig. 26 as input, and converts the cause part that has "increase in voids" and "increase in water supply flow rate" in the result part, respectively, to the cause part.
Search in result database 8. If no new cause is found in this search, the recursive call to the cause determination function 3 ends. In this case, for the former case, we will deal with the "recirculation pump shaft sticking" shown in Figure 27.
21A is searched, and for the latter, "water supply control system abnormality" 21A shown in FIG. 28 is searched.

Next, the status grasping function 2
Each plant state 21A shown in FIG. 8 is input, and all results caused by each state are determined by forward inference based on the cause/effect database 8. As a result, in addition to "increase in voids", a "decreased jet pump flow rate" is obtained as a result of "recirculation pump shaft sticking", and in addition to "increase in water supply flow rate", as a result of "water supply control system abnormality", a "flow rate mismatch" is obtained. is obtained, and as a result of "increase in voids" and "increase in water supply flow rate," both "reactor water level rises" are obtained.

Both plant states with these added are input to the prediction function 15. The prediction function 15 predicts the transition of each plant state, adds "reactor water level = 7" to each plant state as in the previous time, and
The results are input to the consistency checking function 16. The consistency checking function 16 determines that there is no contradiction in both plant states and outputs them as they are. Through the above processing, respective plant states 21D shown in FIGS. 27 and 28 are obtained. The cause enumeration function 13 is activated again, but since no further cause is found, the recursive call to the cause determination function is terminated.

Next, the determination function 18 shown in FIG. 14 is activated, and several plant states obtained at the end of the recursive call of the cause determination function described above, that is, various causes and their predicted results, are used as the reference plant state. It is compared with 21. In this case, the plant state 21D shown in FIGS. 27 and 28 is
This is compared with the plant state 21 shown in FIG. As a result of this comparison, when the cause is ``recirculation pump shaft sticking'', the degree of agreement with plant condition 21 is higher than when the cause is ``abnormality in the water supply control system'' in terms of including ``reduced jet pump flow rate.'' is large. Therefore, "recirculation pump shaft sticking" is determined to be the cause, the plant status 21D shown in FIG. 27 is output as the output 22, and the process by the cause determination function 3 is completed.

Next, the optimal operation determination function 4 is activated. As shown in FIG. 15, the optimal operation determination function 4 inputs the plant status 22 output from the cause determination function 3, and repeatedly executes the operation enumeration function 14, prediction function 15, and status understanding function 2 in this order. is activated to find out all possible operations, and from among them, the decision function 17 selects the operation deemed most appropriate.

First, the operation enumeration function 14 shown in FIG. 16 is activated. That is, the condition part of the operation database 12 (FIG. 5) is searched for each element state of the plant state 22, and if there is a matching condition part, the operation plan corresponding to that condition part is read out, and these operation plans are used. A plant state 22A to which is added is output. In this case, 21D in FIG. 27 is the input, and no operation plan is found for it. Therefore, the operation plan is "do nothing," but in this embodiment, this is expressed by adding no operation plan, and the input plant state (21D in FIG. 27) is output as is.

Next, the prediction function 15 operates by inputting this plant state 22A. ), referring to the transition prediction database 11, calculates the change time of each element state for which prediction calculation of the plant state 22A is possible, determines the minimum change time among them, and then calculates the elapsed time of this minimum change time. Calculates each element state later and outputs the plant state to which the calculated element state is added 22
Output as B. In this example, regarding the reactor water level in the input plant state (21D in Figure 27), "reactor water level = L8" is calculated as the state after the minimum change time has elapsed from the state of "reactor water level = L7". , the plant state with this added (not shown)
is output.

This plant status 22B is input to the status grasping function 2. As mentioned above, the status grasp function 2 is
Execute forward inference based on the cause-and-effect database 8, find the result "turbine trip" for "reactor water level = L8", and further find the results "scrum", "busbar switching", and "turbine trip" for this "turbine trip". Obtain "furnace pressure increase". As a result, a plant state (FIG. 29) to which these new element states have been added is obtained as the output 22C of the state grasping function 2.

Therefore, in order to search for the next necessary operation, the optimal operation determination function 4 is called recursively.

In this second operation search process, the operation enumeration function 14 inputs the plant state 22C obtained from the state grasping function 2 in the first process, and again obtains an operation plan for this from the operation database 12. In this example, the plant status shown in FIG. 29 is input, and the operation "motor-driven water supply pump trip" corresponding to "reactor water level rise, reactor water level = L8" is obtained. Also, apart from this, "do nothing"
is automatically selected as a possible operation. Then, each plant state to which these operation plans have been added, that is, the plant state 22A shown in FIGS. 30 and 31, is calculated by the prediction function 15.
is input.

The prediction function 15 predicts the transition of the plant state when each of the retrieved operations is performed using the procedure described above. The state after the minimum change time has elapsed is further predicted from the state after the elapse of the minimum change time. In this example, when performing a "motor-driven water supply pump trip" (3rd
Figure 0) shows "reactor pressure rise" and "reactor water level sudden drop,"
It is predicted that the reactor water level will be “L7”, and
Even in the case of "doing nothing" (FIG. 31), it is predicted that "reactor pressure will rise" and "reactor water level will fall, reactor water level = L7". The plant state with these states added is input to the state grasping function 2.

The state understanding function 2 determines the result brought about by each element state in the input plant state by forward reasoning based on the cause/effect database 8.
That is, in the case of the operation plan ``motor-driven water supply pump trip'' (Figure 30), ``reactor water level drop'' is found as a result for the cause ``motor-driven water supply pump trip'', and this is added to the plant in Figure 30. State 22C is obtained. Further, in the case of the operation plan "Do nothing" (FIG. 31), no result is found, and as a result, the plant state 22C in FIG. 31 is obtained.

It should be noted that when different states for the same attribute are added at different processing stages, only the states relevant to the following explanation are shown in the figure for the sake of clarity.

Subsequently, the optimal operation determining function 4 is called again recursively to search for the next required operation. In other words, the plant status 22C in FIGS. 30 and 31 is changed to the operation enumeration function 1.
4 is input. However, in this case, there are no new operation plans to be added for any plant status.

Next, the prediction function 15 receives these plant states as input 22A and predicts the state after the elapse of the minimum change time of each element state that can be predicted and calculated. As a result, the plant state (Fig. 30, 22
(same as C), it is predicted that "reactor pressure rises, high" and "reactor water level drops rapidly, L = 4", so by adding these, plant state 2 in Fig. 32 is predicted.
2B is obtained, and from the plant state (same as Fig. 31 22C) brought about by the operation plan "Do nothing", "reactor pressure rises, high" and "reactor water level falls, L=
4'', so by adding them, the plant state 22B in FIG. 33 is obtained.

These plant states are input to the state grasping function 2. The status understanding function 2 analyzes the causes and results of each element status in the input plant status.
It is determined by forward inference based on the result database 8. As a result, regarding the plant status brought about by the operation plan "motor-driven water supply pump trip", the cause is "bypass valve open" for the cause "furnace pressure high", and the cause is "scram (predetermined time elapsed)" (Note:
The result ``void reduction'' for the cause (two minimum change times have elapsed after the scram) is found, and the result ``reactor water level fall'' for the cause ``void reduction'' is found, and these are added. As a result, a plant state 22C in FIG. 32 is obtained. Furthermore, the same result as above is found for the plant state brought about by the operation plan "Do nothing", and the plant state 22C in FIG. 33 is obtained.

The recursive call is performed a predetermined number of times (in this example, 3
It will be stopped after the process (time) is completed.

Next, the decision function 17 shown in FIG. 17 executes the plant state (22 in FIGS.
C), determine the operation that will result in the most desirable plant conditions. In other words, according to the input plant statuses, when the operation ``motor-driven water supply pump trip'' is executed, ``the reactor water level suddenly drops''.
occurs, and it turns out that the reactor water level drops more rapidly compared to the "reactor water level fall" that would occur if the operation "do nothing" was performed. Therefore, the operation "do not lower the reactor water level too much for safety reasons" In light of the principles, it is determined that ``do nothing'' is the most appropriate course of action for the presumed cause, ``recirculation pump shaft stuck.'' As a result, from the optimal operation determination function 4,
The plant state 22C in FIG. 33 is output as the output plant state 23.

Next, using this plant state 23 as input, the detailing function 5 shown in FIG. 18 is activated. In this example, since the optimal operation is "do nothing", there is no information added by the detailing function 5, and the input plant state is output as is.

However, if the selected operation plan includes activation of the high-pressure water injection system (for example, if the above-mentioned optimal operation determination function 4 is further recursively called, the reactor water level will eventually rise to L2). As a corresponding operation plan, "high-pressure water injection system" is retrieved from the operation database 12 (FIG. 5).) Then, the detailing function 5 retrieves the detailed operation of the corresponding high-pressure water injection system from the detailed database 9. If the operating conditions are satisfied, the plant status 24 to which this detailed operating method is added is output.

On the other hand, if the operating conditions are not met, the detailing function 5 updates the plant status 2 with the reason for rejection: "The high-pressure water injection system cannot be used."
5 and restart the optimal operation determination function 4.
Have the operator plan the operation again (see Figure 2).

Next, as shown in FIG. 2, the similar case search function 6 shown in FIG. 19 is activated by inputting the plant status 24 from the detailing function 5. The similar case search function 6 searches a case database 10 containing preliminary analysis results and past actual cases, as shown in FIG. Search for examples. In this case, the case of recirculation pump shaft sticking recorded as case number 1 is searched, and the case number is added to the plant state 24 to form the plant state 26, which is output from the similar case search function 6. be done.

The plant state 26 obtained in this way is
The information is input to the guidance creation function 7 shown in FIG. 20, converted for output to a display device (for example, into a character code and format for CRT), and output. As a result of the above-mentioned series of processes, this plant state includes the determined cause, the element state that should occur as a result, the operation that is considered optimal for dealing with it, and the element state that should occur as a result of this operation. It includes detailed operating methods and similar examples.

The output of the guidance creation function 7 is given to the display device (for example, CRT) 71 shown in FIG.
It will be displayed there. The plant operator can view the displayed content and operate the plant via the control panel 70 in accordance with the displayed content.

The process described above is performed by generating a new alarm,
By re-executing in response to an operator's request or an interrupt using the internal clock of the driving guide device,
It is possible to constantly provide guidance to operators in response to new situations.

Many variations in embodiments of the invention are possible, as described below. For example, when inputting plant data, the status understanding function 2 is activated,
It may be input for each element state at the time when the cause/effect database 8 is actually used, such as when it is referenced.

In data exchange function 1, when multiple measured values are given for one process quantity, the decision logic for narrowing down to one is not to select the most probable value by majority vote, but is preferable for the plant. You may intentionally choose a value that does not exist.

The cause determination function 3 may output all the found causes and perform the subsequent processing for each cause, without determining only one cause, unless there is a contradiction.

In the optimum operation determination function 4, instead of determining only one operation, all operations that match the driving purpose may be output, and the operator may select one of them. Alternatively, instead of exploring all possible operations,
The process may be terminated and output when a preset number of items matching the driving purpose have been retrieved.

The operation database 12 may be integrated with the cause/effect database 8, and a code indicating which data it belongs to may be attached to each data item so that the data can be distinguished.

The detailing function 5 and similar case search function 6 may be activated only after receiving an instruction from an operator. Regarding the execution order, the similar case search function may be executed first, or both may be executed simultaneously. Alternatively, one can be omitted.

Instead of storing the algorithms themselves that can be directly interpreted and executed by the prediction function 15, the transition prediction database 11 may store information for calling calculation subroutines that the prediction function 15 should call and execute.

For controlling the cause determination function 3 and the optimum operation determination function 4, similar control may be performed using a stack instead of the re-call method. Alternatively, the cause determination function and the optimal operation determination function can be realized by serially connecting the required number of hardware that performs these functions.

Prediction function 1 when determining the cause and deciding on the operation
5 works, it is possible to efficiently determine the most probable cause in cause estimation,
In addition, when making operational decisions, we can not only suggest safe operating methods, but also the optimal operating methods, even for phenomena that are unlikely to actually occur, such as a stuck recirculation pump shaft. be done.

Abnormality cause estimation devices and driving guide devices that simply use CCT and conventional knowledge engineering methods are capable of determining the most probable cause, and in dealing with accidents that are unlikely to occur in practice, such as those mentioned above. Outputting guidance on safe operation methods
Although it is not completely impossible, making it possible would require a complex and large-scale database, which would be difficult to create and maintain. In other words, there are many cases in which multiple causes can be considered for a certain abnormal state in a plant indicated by measurement data, but regardless of which conventional method is used, it is not possible to trace these multiple causes and their results, respectively. It is necessary to prepare data that expresses causal relationships as much as possible, and
In order to provide a device with the ability to predict changes in plant conditions, causal relationship data that includes time changes in plant conditions must be prepared. Therefore, the amount of data in the database must be enormous.

In addition, the contents of each database are shown in Figures 3 to 7.
As shown in the figure, each constituent unit is independent, so even in extreme cases such as a phenomenon not included in the database occurring, operators can If only the characteristics of a phenomenon are entered into a database as a plant state, it is expected that the equipment will be able to respond based on this by using these existing databases, and in that sense, the flexibility or expandability of the system will be improved. improves.

〔Effect of the invention〕

As is clear from the above explanation, according to the present invention, without asking the operator a question in the middle of reasoning,
Since inferences are automatically performed online, the cause of abnormal conditions in the target equipment can be estimated extremely quickly and without the risk of being affected by errors in judgment by the operator. Automatically provides operational guidance to address the cause.

In addition, as a result of data-driven inference processing,
Online processing can be performed efficiently.

Furthermore, by predicting the element state that will be reached after an operation is executed, it is possible to prevent incorrect operation guidance from being issued and ensure safety.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the overall configuration of a boiling water reactor plant to which an embodiment of the present invention is applied. FIG. 2 is a functional block diagram of an embodiment of the operation guide device of the present invention implemented on the computer shown in FIG. FIG. 3 is a diagram showing an example of the contents of the cause/effect database. FIG. 4 is a diagram showing an example of the contents of the transition prediction database. FIG. 5 is a diagram showing an example of the contents of the operation database. FIG. 6 is a diagram showing an example of the contents of the detailed database. FIG. 7 is a diagram showing an example of the contents of the case database. FIG. 8 is a flowchart of the data conversion function. FIG. 9 is a flowchart of the state grasping function.
FIG. 10 is an overall functional block diagram of the cause determination function. FIG. 11 is a flowchart of the cause enumeration function. FIG. 12 is a flowchart of the prediction function. FIG. 13 is a flowchart of the consistency check function. FIG. 14 is a flowchart of the determination function. FIG. 15 is an overall functional block diagram of the optimum operation determining function. FIG. 16 is a flowchart of the operation enumeration function. 1st
FIG. 7 is a flowchart of the decision function. 1st
FIG. 8 is a flowchart of the detailing function. FIG. 19 is a flowchart of the similar case search function. FIG. 20 is a flowchart of the guidance creation function. FIG. 21 is a diagram showing an example of the contents of plant status data output from the data conversion function. FIG. 22 is a diagram showing an example of the contents of plant status data output from the status grasping function. FIGS. 23 and 24 are diagrams showing examples of the contents of plant status data output from the cause enumeration function. FIGS. 25 and 26 are diagrams showing examples of the contents of plant state data output from the prediction function of the cause determination function. Figures 27 and 28 show plant status data output from the cause enumeration function and plant status data output via the status grasping function and prediction function in processing by the recursively called cause determination function. It is a figure which shows an example. FIG. 29 is a diagram showing an example of the contents of plant state data output from the state grasping function within the optimum operation determining function. Figures 30 and 31 show plant status data output from the operation enumeration function and plant status data output from the prediction function and status grasping function during processing by the recursively called optimal operation determination function. It is a figure showing an example of the contents. Figures 32 and 33 show the plant status data output from the prediction function and the plant status data output via the status grasping function during processing by the optimal operation determination function that is recursively called for the second time. It is a figure showing an example of contents. 2...Status understanding function, 3...Cause determination function, 4
...Optimum operation determination function, 5...Detailing function, 6...
... Similar case search function, 8... Cause/effect database, 9... Detailed database, 10... Case database, 11... Transition prediction database, 1
2... Operation database, 13... Cause enumeration function, 14... Operation enumeration function, 15... Prediction function,
16... Consistency confirmation function, 17... Operation determination function, 18... Cause final determination function, 51... Nuclear reactor, 68... Electronic computer, 68A... Input device,
68B...Central processing unit, 68C...Storage device,
71...Display device, 77...Processing program.

Claims (1)

[Claims] 1. An input device connected to a detection device that detects and outputs the state of a target device, a storage device that stores required information, and An operation that uses an electronic computer having a processing device that performs inference according to a predetermined procedure to execute a process that includes the following steps to create an operation guide for dealing with the abnormal state of the target device. Guide method. (1) Creating at least one element state data representing a partial state of the target device in a predetermined format suitable for rule-based inference processing from the output of the detection device input to the input device (2) ) searching for a rule corresponding to the created element state data from a plurality of rules stored in the storage device, each of which associates the element state of the target device as a pair of a cause and its effect;
Furthermore, various rules that form a chain of causal relationships from this rule are searched, the cause forming the end of the chain is extracted, and at least one cause of the abnormal state of the target device is identified from among the extracted causes. Selecting and estimating the cause step (3) Dealing with the selected cause using countermeasure operation information stored in the storage device that indicates operations for dealing with multiple causes in association with those causes. The stage of deciding what operations to do. 2. An operation guide method according to claim 1, wherein the element state data creation step creates only element state data indicating an abnormal state of the target device. 3. An operation guide method according to claim 1 or 2, wherein the step of estimating the cause includes the following steps. (1) Starting from the created element state data and performing forward inference using the plurality of rules; (2) Adding the element state obtained by the forward inference to the created element state data. (3) Performing backward inference using the plurality of rules based on each element state resulting from the addition. 4. In the method recited in any one of claims 1 to 3, the cause estimating step includes, at a certain stage during the process, determining the state of a certain element stored in the storage device. An operation guide method, comprising the step of estimating an element state that is expected to be reached subsequently due to causes estimated up to that point, according to an estimation procedure for estimating an element state that is expected to be reached. 5 In the method recited in claim 4, the cause estimation step further includes comparing the estimated element state that should be reached later with the created element state data, and determining the difference between them. An operation guide method comprising the step of selecting the causes estimated up to that point based on the presence or absence of a contradiction. 6. An operation guide method according to claim 5, wherein the cause estimation step further includes the following steps. (1) The step of performing backward inference again using the plurality of rules based on the element states of each cause remaining as a result of the selection, to find out the causes that can bring about each element state (2) The step ( (3) estimating the element state that is expected to be reached later due to each cause found in step (1) using at least the estimation procedure; (3) estimating the element state estimated in step (2) using the created element state; A step (4) of selecting the causes found in step (1) based on the presence or absence of inconsistency between the element state data and the element state data. (4) Recursively performing steps (1) to (3) Iterative stages. 7. In the method recited in any one of claims 4 to 6, the step of estimating the element state that is to be reached after that involves determining the element state to be estimated by the plurality of rules. An operational guidance method comprising a step of determining by forward reasoning using . 8. In the method set forth in claims 5 to 7, the cause estimation step includes determining, for each of the plurality of causes selected as having no contradiction, the estimated The operation determination step includes the step of comparing the element state with the created element state data and selecting one of the plurality of causes based on the degree of matching between them, and the operation determining step An operational guide that explores operations to deal with the situation. 9. In the method according to any one of claims 1 to 8, the operation determining step determines the element state that is to be reached after executing the operation retrieved from the countermeasure operation information. An operational guidance method comprising the step of predicting according to an estimation procedure. 10 In the method set forth in claim 9, the operation determining step includes a step of searching for an operation to deal with the predicted element state using the coping operation information, and a step of searching for an operation to deal with the predicted element state; An operation guide method comprising the steps of: predicting an element state that will be reached after the operation is executed according to the estimation procedure; and recursively repeating both steps. 11. In the method described in claim 9 or 10, the prediction step includes predicting the resulting element state caused by the element state obtained by the prediction using the plurality of rules. An operational guidance method that includes a step of predicting by inference. 12 In the method described in claims 9 to 11, the operation determining step includes determining the predicted result of each of the retrieved operations or sets of operations. An operation guidance method comprising the steps of evaluating element conditions against predetermined operating principles and determining optimal operation based on the results.